Model-driven software development uses abstract models as central artifacts, but concepts for modularization and reuse are largely lacking. This dissertation presents a formally grounded approach to the composition of UML class diagrams and defines the denotational semantics of the language profile CD4Analysis. The composition operator detects conflicts and checks the compatibility of models, and a tool called CDMerge has been developed for the Language Workbench Monticore. Additionally, Modelpedia is described as a web-based repository for quality-assured versioning and provision of models and tools. A case study demonstrates the practical application of the methods as well as the advantages of modular data models for reuse, efficiency, and agile development processes.