The integration of safety analysis into the early stages of system design using Model-Based Systems Engineering (MBSE) is critical for developing trustworthy safety-critical systems. Methodologies like SafeDeML provide a semi-formal way to embed safety artifacts such as faults, failures, and diagnoses directly within SysML models. However, a significant gap remains in translating these annotated models into a formal domain for rigorous, automated analysis of system-wide failure propagation. This thesis proposes the development of a methodology and a corresponding toolchain to automatically derive a formal, compositional analysis model from a SysML system architecture annotated with SafeDeML. The goal is to enable the automated verification of system-level safety goals against failure propagation scenarios originating from component-level faults.