A modeling process is presented for extracting timing-accurate simulation models from complex embedded real-time systems. The process is supported by two complementary methods for tool-supported model extraction, Model Synthesis and Hybrid Model Extraction. The generated models enable impact analysis for complex real-time systems with respect to dynamic system properties, such as timing and resource usage. This can make software mainetnance more predictable with respect to time-to-market and development costs, since timing errors can be identified early and avoided. The contribution of the paper is the modeling process, the Hybrid Model Extraction method and an interactive modeling tool, MASS, designed to support Hybrid Model Extraction of large implementations in C.