One of the main challenges in the design of real-time systems is how to derive correct and efficient implementations from platform-independent specifications.
We present a general implementation method in which the application is represented by an
model consisting of a set of interacting components. The abstract model executes sequentially components interactions atomically and instantaneously. We transform abstract models into
models representing their execution on a platform. Physical models take into account execution times of interactions and allow their parallel execution. They are obtained by breaking atomicity of interactions using a notion of partial state. We provide safety conditions guaranteeing that the semantics of abstract models is preserved by physical models. These provide bases for implementing a parallel execution engine coordinating the execution of the components. The implementation has been validated on a real robotic application. Benchmarks show net improvement of its performance compared to a sequential implementation.