This thesis presents a thorough and significant work on the properties of radio sources as derived from deep 610-MHz GMRT data and ancillary multi-wavelength data. The faint radio sources at 610-MHz are found out to distances such that the objects are seen as they were when the universe was less than half its current age. These data provide a first look at the faint radio sky at sensitivities that will soon be achieved by key programs on the South African MeerKAT radio telescope, and thus take a first step in the exploration of the radio universe that will be made by the Square Kilometre Array. I report deep 610-MHz GMRT observations of the EN1 field, a region of 1.86 deg2 . We achieve a nominal sensitivity of 7.1µ Jy beam−1 . From our 610 MHz mosaic image, we recover 4290 sources after accounting for multiple component sources down to a 5σ flux density limit of 35.5 µ Jy. From this data, I derive the 610 MHz source counts applying corrections for completeness, resolution bias and Eddington bias. The 610- MHz source counts show a flattening at flux densities below 1 mJy. The source counts are higher than previous observations at this frequency below this break. However, they are generally consistent with recent models of the low-frequency source population. Using ancillary multi-wavelength data in the field, I investigate the key issue of source population classification using the deepest data at an intermediate-low frequency (higher than LOFAR and lower than JVLA), where previous work has not been sensitive enough to reach the µJy population. By cross-matching against the multi-wavelength data, I identify 72% of the radio sample having reliable redshifts, of which 19% of the redshifts are based on spectroscopy. From the classification, I obtain 1685 sources as Star-Forming Galaxies (SFGs), 281 sources Radio-Quiet (RQ) and 339 sources Radio-Loud (RL) Active Galactic Nuclei (AGN) for the sub-sample with redshifts and at least one multi-wavelength AGN diagnostic. SFGs are mostly low-power radio sources, i.e L610 MHz < 1025 W Hz−1 while RQ AGN and RL AGN have radio powers L610 MHz > 1025 W Hz−1 . From cross-matching my sample with other radio surveys (GMRT at 325-MHz, FIRST at 1.4-GHz and JVLA at 5-GHz), I obtain the median spectral index from 325-MHz to 610-MHz to be −0.80 ± 0.29, 610-MHz to 1.4-GHz to be −0.83 ± 0.31 and 1.4-GHz to 5-GHz to be −1.12 ± 0.15. The main result is that the median spectral index appears to steepen at the highest frequency. With the above catalogue in hand, I use the non-parametric V/Vmax test and the radio luminosity function to investigate the cosmic evolution of different source populations. I study SFGs and derive their IR-radio correlation and luminosity function as a function of redshift. By integrating the evolving SFG luminosity functions I also derive the cosmic star formation rate density out to z = 1.5. I address the long standing question about the origin of radio emission in RQ AGN. I compare the star formation rate (SFR) derived from their far-infrared luminosity, as traced by Herschel, with the SFR computed from their radio emission. I find evidence that the main contribution to the radio emission of RQ AGN is the star formation activity in their host galaxies. At high luminosities, however, both SFGs and 1 RQ AGN display a radio excess when comparing radio and infrared star formation rates. The vast majority of our sample lie along the SFR − M? ”main sequence” at all redshifts when using infrared star formation rates. This result opens the possibility of using the radio band to estimate the SFR even in the hosts of bright AGN where the optical-to-mid-infrared emission can be dominated by the AGN. I investigate the evolution of radio AGN out to z ∼ 1.5 with continuous models of pure density and pure luminosity evolution with Φ? ∝ ( 1 + z)(2.25±0.38)−(0.63±0.35)z and L610 MHz ∝ ( 1 + z)(3.45±0.53)−(0.55±0.29)z respectively. I also constrain the evolution of RQ AGN and RL AGN separately with a continuous model of pure luminosity evolution. For the RQ and RL AGN, we find a fairly mild evolution with redshift best fitted by pure luminosity evolution with L610 MHz ∝ ( 1 + z)(2.81±0.43)−(0.57±0.30)z for RQ AGN and L610 MHz ∝ ( 1 + z)(3.58±0.54)−(0.56±0.29)z for RL AGN. The results reveal that the 610 MHz radio AGN population thus comprises two differently evolving populations whose radio emission is mostly SF-driven or AGN-driven respectively. Finally, I probe the infrared-radio correlation and radio spectral indices of the faint radio population using stacking. I stack infrared sources in the EN1 field using the MIPS 24 micron mid-infrared survey and radio surveys created at 325 MHz, 610 MHz and 1.4 GHz. The stacking experiment shows a variation in the absolute strength of the infrared-radio correlation between these three different frequencies and the MIPS 24 micron band. I find tentative evidence of a small deviation from the correlation at the faintest infrared flux densities. The stacked radio spectral index analyses reveal that the majority of the median stacked sources exhibit steep spectra, with a spectral index that steepens with frequency between α 325 610 and α 610 1400. This work is particularly useful to pave the way for upcoming radio surveys with SKA pathfinders and precursors.