More and more devices of our everyday life are computerized with smart embedded systems and software-intensive electronics. Whenever these pervasive embedded systems interact with the physical world and have the potential to endanger human lives or to cause significant damage, they are considered safety-critical. To avoid any unreasonable risk originating from the failure of such systems, stringent development processes, safety engineering practices, and safety standards are followed and applied for their development and operation. Thereby, functional safety mechanisms provide technical solutions to detect faults or control failures in order to achieve or maintain a safe state. In consequence, the requirements on their dependable and trustworthy operation are correspondingly high. On this background, this thesis is concerned with the assessment and enhancement of functional safety mechanisms in software-intensive safety-critical embedded systems at the example of automotive systems based on the AUTOSAR standard. An established technique for dependability assessments is fault injection (FI). The effective adaptation and application of FI to modern embedded safety-critical software systems, such as AUTOSAR, is non-trivial due to their complexity and multiple levels of abstraction that are introduced by model-based development, layered architectures, and the integration of components from various suppliers, which impact the overall customizability, usability, and effectiveness of experiments. Facing these challenges, this thesis develops a complete FI process, which includes a guidance framework for the systematic and automated instrumentation with FI test code, a FI framework for the automated execution of experiments, a detailed discussion on the derivation of fault models, and the demonstration of their effective application in two case studies that uncovered an actual deficiency in a functional safety mechanism. Due to the high cost-saving potential, functionality of varied criticality is increasingly integrated into so-called mixed-criticality systems. To provide efficient protection of critical tasks, functional safety mechanisms benefit from accounting for different criticality levels. At the example of AUTOSAR's timing protection, we illustrate the issues emerging from the lack of criticality awareness and the resulting indirect protection of critical tasks. As mitigation, we propose a novel monitoring scheme that directly protects critical tasks by providing them with execution time guarantees and implement our approach as an enhancement to the existing monitoring infrastructure.