Cyber-physical systems (CPS) implement different kinds of controlling and monitoring technologies via the integration of hardware, software and complex communication networks. Such systems are usually designed by using model-driven approaches and tools. However, validation and performance-testing of these tools are not easy because of the lack of publicly available scalable set of benchmark models. Making such models manually for testing would be time consuming and expensive, but by automatic model generation, we can overcome this problem, increasing productivity while reducing production costs. I have implemented and documented a framework that is capable of generating realistic models from a defined domain based on real system properties. By examining these properties of a real system, I was also able to define a method for calculating the realisticness of models. Instances generated via this method proved to have similar metrics to real world system models. These metrics and other performance related behavior have been measured and visualized. The framework currently generates smart grid (a kind of CPS) based instance models, but it could be extended to operate according to a differently domain.