The number of embedded controllers in charge of physical systems has rapidly increased over the past years. Embedded controllers are present in every aspect of our lives, from our homes to our vehicles and factories. The complexity of these systems is also more than ever. These systems are expected to deliver many features and high performance without trading off in robustness and assurance. As systems increase in complexity, however, the cost of formally verifying their correctness and eliminating security vulnerabilities can quickly explode. On top of the unintentional bugs and problems, malicious attacks on cyber-physical systems (CPS) can also lead to adverse outcomes on physical plants. Some of the recent attacks on CPS are focused on causing physical damage to the plants or the environment. Such intruders make their way into the system using cyber exploits but then initiate actions that can destabilize and even damage the underlying (physical) systems.Given the reality mentioned above and the reliability standards of the industry, there is a need to embrace new CPS design paradigms where faults and security vulnerabilities are the norms rather than an anomaly. Such imperfections must be assumed to exist in every system and component unless it is formally verified and scanned. Faults and vulnerabilities should be safely handled and the CPS must be able to recover from them at run-time. Our goal in this work is to introduce and investigate a few designs compatible with this paradigm. The architectures and techniques proposed in this dissertation do not rely on the testing and complete system verification. Instead, they enforce safety at the highest level of the system and extend guaranteed safety from a few certified components to the entire system. These solutions are carefully curated to utilize unverified components and provide guaranteed performance.