Model-to-model (M2M) transformation is a key ingredient in a typical Model-Driven Engineering workflow and there are several tailored high-level interpreted languages for capturing and executing such transformations. While these languages enable the specification of concise transformations through task-specific constructs (rules/mappings, bindings), their use can pose scalability challenges when it comes to very large models. In this paper, we present an architecture for optimising the execution of model-to-model transformations written in such a language, by leveraging static analysis and automated program rewriting techniques. We demonstrate how static analysis and dependency information between rules can be used to reduce the size of the transformation trace and to optimise certain classes of transformations. Finally, we detail the performance benefits that can be delivered by this form of optimisation, through a
series of benchmarks performed with an existing transformation language (Epsilon Transformation Language - ETL) and EMF-based models. Our experiments have shown considerable performance improvements compared to the existing ETL execution engine, without sacrificing any features of the language.