This thesis is concerned with studies on spatially developing high speed mixing layers with twin objectives: (a) to provide enhanced and detailed understanding of spatial development of two-dimensional mixing layer emanating from splitter plate through large eddy simulation (LES, from now on) technique and (b) to evolve a consistent strategy for Unsteady Reynolds Averaged Navier-Stokes (URANS) approach to mixing layer calculations. The inspiration for this work arose out of the explanations that were being developed for the reduction in the mixing layer thickness with compressibility (measured by a parameter called convective Mach number, Mc). The reasons centered around increased stability, increase in compressible dissipation that was later discounted in favor of reduction in production and pressure-strain terms (with Mc, of course). These were obtained with direct numerical simulations (DNS) or LES techniques with homogeneous shear flow or temporal mixing layer. As apart, there was also a wide held view that using RANS (steady) techniques did not capture the compressibility effects when used in a way described above and so classical industrial codes for computing mixing- layer-embedded flows are unsuitable for such applications. Other important aspects that come out of the examination of literature are: the mixing layer growth is controlled in the initial stages by the double- boundary layer profile over the splitter plate and results in the mixing layer growth that is somewhat irregular due to doubling and merging of vertical structures. The view point of a smooth growth of the mixing layer is a theo- retical approximation arising out of the use of a smooth tan-hyperbolic profile that results at larger distances from the splitter plate. For all practical applications, it is inferred that the initial development is what is important because the processes of ignition and stable combustion occur close to the splitter plate. For these reasons, it was thought that understanding the development of the mixing layer is best dealt with using accurate spatial simulation with the appropriate initial profile. The LES technique used here is drawn from an OpenFOAM approach for dissimilar gases and uses one-equation Eddy Model for SGS stresses. The temporal discretization is second order accurate backward Euler and spatial discretization is fourth order least squares; the algorithm used for solving the equations is PISO and the parallelized code uses domain decomposition approach to cover large spatial domain. The calculations are performed with boundary layer profiles over the splitter plate and an initial velocity field with white noise-like fluctuations to simulate the turbulence as in the experiments. Grid independence studies are performed and several experimental cases are considered for comparison with measured data on the velocity and temperature fields as well as turbulent statistics. These comparisons are excellent for the mean field behavior and moderately acceptable for turbulent kinetic energy and shear stress. To further benefit from the LES approach, the details of the mixing layer are calculated as a function of four independent parameters on which the growth depends: convective Mach number (Mc = (U1 -U2)/ (a1 +a2)), stream speed ratio (r = U2=U1), stream density ratio (s = p2/p1) and the average velocity of the two streams ((U1+U2)=2) and examine the various terms in the equations to enable answering the questions discussed earlier. It is uncovered that r has significant influence on the attainment of self similarity (which also implies on the rate of removal of velocity defect in the double-boundary layer profile) and other parameters have a very weak influence. The minimum velocity variation with distance from the splitter plate has the 1/paxial distance behavior like in wakes; however, after a distance, departure to linear rise occurs and the distance it takes for this to appear is delayed with Mc. Other features such as the coherent structures, their merger or break up, the area of the structures, convective velocity information extraction from the coherent structures, the behavior of the pressure field in the mixing layer through the field are elucidated in detail; the behavior of the correlations between parameters (like pressure, velocity etc) at different points is used to elucidate the coherence of their fluctuating field. The effects of the parameters on the energy spectra have expected trends. An examination of the kinetic energy budget terms reveals that • the production term is the main source of the xx turbulence stress, whereas it is not significant in the yy component. • A substantial portion of this is carried by the pressure-velocity coupling from the xx direction to the yy direction, which becomes the main source term in the yy component. • Both, the production term as well as the pressure-velocity term show a clear decrease with increase in Mc. The high point of the thesis is related to using the understanding derived from an analysis of various source terms in the kinetic energy balance to evolve an unsteady Reynolds Averaged Navier Stokes (URANS) model for calculating high speed mixing layers, a subject that has eluded international research till now. It recognizes that the key feature affected by ompressibility is related to the anisotropy of the stress tensor. The relationship between stress component (_Txy) and the velocity gradient (Sxy) as obtained from LES is set out in the form of a simple relationship accounting for the effects of other parameters obtained earlier in this thesis. A minor influence due to _Tyy is extracted by describing its dependence on Sxy again as gleaned from LES studies. The needed variation of Prandtl and Schmidt numbers through the field is extracted. While the detailed variations can in fact be taken into account in URANS simulations, a simple assumption of these values being around 0.3 is chosen for the present simulations of URANS. Introduction of these features into the momentum equation gives the much expected variation of the reduction in the growth rate of the mixing layer with convective Mach number as in experiments. The relationships that can be used in high speed mixing layers are Introduction of these features into the momentum equation gives the much expected variation of the reduction in the growth rate of the mixing layer with convective Mach number as in experiments. This is then a suggested new approach to solve high speed mixing layers. While it can be thought that the principal contributions of the thesis are complete here, an additional segment is presented related to entropy view of the mixing layer. This study that considers the mixing layer with two different species expresses various terms involved in the entropy conservation equation and obtains the contribution of various terms on the entropy change for various Mc. It is first verified that the entropy derived from the conservation equation matches with those calculated from fluid properties, entropy being a state variable. It is shown that irreversible diffusion comes down the most with convective Mach number. Left: This image shows pictorially the flow of source of turbulent stress from the axial to the cross wise turbulent stress. Production (Σ) of turbulence happens mainly in the xx direction, a part of it is carried by the pressure-velocity correlation to the yy direction, which itself has a low production. With increasing Mc, both the production as well as the pressure-velocity correlation decrease. Right: This image shows the growth rate obtained from simulations scaled with the incompressible growth rate, of LES and RANS in the background of experiments (others). As is clear, the growth rate obtained is well within the band of experimental results.