Developing complex software systems is costly, time-consuming and error-prone. Model- driven development (MDD) promises to improve software productivity, timeliness, quality and cost through the transformation of abstract application models to code-level implementations. However, it remains unreasonably difficult to build the modeling languages and translators required for software synthesis. This difficulty, in turns, limits the applicability of MDD, and makes it hard to achieve reliability in MDD tools. This dissertation research seeks to reduce the cost, broaden the applicability, and increase the quality of model-driven development systems by embedding modeling languages within established formal languages and by using the analyzers provided with such languages for synthesis purposes to reduce the need for hand coding of translators. This dissertation, in particular, explores the proposed approach using relational logic as expressed in Alloy as the general specification language, and the Alloy Analyzer as the general-purpose analyzer. Synthesis is thus driven by finite-domain constraint satisfaction. One important aspect of this work is its focus on partial specifications of particular aspects of the system, such as application architectures and target platforms, and synthesis of partial code bases from such specifications. Contributions of this work include novel insights, methods and tools for (1) synthesizing architectural models from abstract application models; (2) synthesizing partial, platform-specific application frameworks from application architectures; and (3) synthesizing object-relational mapping tradeoff spaces and database schemas for database-backed object-oriented applications.