Discrete-event stochastic models are a widely used approach for studying the behavior of systems that have not been implemented or that it would be too costly to examine directly. Valuable analysis depends on carefully constructed, well-founded models, which are very difficult for humans to create. To address this problem, we propose a framework for generating detailed, low-level models from high-level, block-diagram-style graphical models. Our approach uses extensible, collaborative ontology libraries that contain information about the types of components in a system, the types of relationships that connect those components, and fragments of low-level models that can be constructed together based on the definition of a high-level system model. This framework has been implemented and used in several case studies. We describe the framework and how model generation works by examining its use to generate complex ADversary VIew Security Evaluation (ADVISE) models.