This thesis addresses software design within the field of Information and Communications Technology for Development (ICTD). Specifically, it makes a case for the design and development of software which is custom-made for the context of marginalised rural areas (MRAs). One of the main aims of any ICTD project is sustainability and such sustainability is particularly difficult in MRAs because of the high costs of projects located there. Most literature on ICTD projects focuses on other factors, such as management, regulations, social and community issues when discussing this issue. Technical matters are often down-played or ignored entirely. This thesis argues that MRAs exhibit unique technical characteristics and that by understanding these characteristics, one can possibly design more cost-effective software. One specific characteristic is described and addressed in this thesis – a characteristic we describe here for the first time and call a network island. Further analysis of the literature generates a picture of a distributed network of access nodes (DANs) within such network islands, which are connected by high speed networks and are able to share resources and stimulate usage of technology by offering a wide range of services. This thesis attempts to design a fitting middleware platform for such a context, which would achieve the following aims: i) allow software developers to create solutions for the context more efficiently (correctly, rapidly); ii) stimulate product managers and business owners to create innovative software products more easily (cost-effectively). A given in the context of this thesis is that the software should use free/libre open source software (FLOSS) – good arguments do also exist for the use of FLOSS. A review of useful FLOSS frameworks is undertaken and several of these are examined in an applied part of the thesis, to see how useful they may be. They form the basis for a walking skeleton implementation of the proposed middleware. The Spring framework is the basis for experiments, along with Spring-Webservices, JMX and PHP 5’s web service capabilities. This thesis builds on three years of work at the Siyakhula Living Lab (SLL), an experimental testbed in a MRA in the Mbashe district of the Eastern Cape of South Africa. Several existing products are deployed at the SLL in the fields of eCommerce, eGovernment and eLearning. Requirements specifications are engineered from a variety of sources, including interviews, mailing lists, the author’s experience as a supervisor at the SLL, and a review of the existing SLL products. Future products are also investigated, as the thesis considers current trends in ICTD. Use cases are also derived and listed. Most of the use cases are concerned with management functions of DANs that can be automated, so that operators of DANs can focus on their core business and not on technology. Using the UML Components methodology, the thesis then proceeds to design a middleware component architecture that is derived from the requirements specification. The process proceeds step-by-step, so that the reader can follow how business rules, operations and interfaces are derived from the use cases. Ultimately, the business rules, interfaces and operations are related to business logic, system interfaces and operations that are situated in specific components. The components in turn are derived from the business information model, that is derived from the business concepts that were initially used to describe the context for the requirements engineering. In this way, a logical method for software design is applied to the problem domain to methodically derive a software design for a middleware solution. The thesis tests the design by considering possible weaknesses in the design. The network aspect is tested by interpolating from formal assumptions about the nature of the context. The data access layer is also identified as a possible bottleneck. We suggest the use of fast indexing methods instead of relational databases to maintain flexibility and efficiency of the data layer. Lessons learned from the exercise are discussed, within the context of the author’s experience in software development teams, as well as in ICTD projects. This synthesis of information leads to warnings about the psychology of middleware development. We note that the ICTD domain is a particularly difficult one with regards to software development as business requirements are not usually clearly formulated and developers do not have the requisite domain knowledge. In conclusion, the core arguments of the thesis are recounted in a bullet form, to lay bare the reasoning behind this work. Novel aspects of the work are also highlighted. They include the description of a network island, and aspects of the DAN middleware requirements engineering and design. Future steps for work based on this thesis are mapped out and open problems relating to this research are touched upon.