"Computer systems have gone through tremendous changes in the past fewdecades. Relatively large general purpose computers dominated the early daysof computers. With time, demand increased for smaller, more dedicated computersystems, called embedded systems. These systems perform a specificset of functions interacting with the physical environment, often in real-time.Real-time embedded systems are found today in many application domainssuch as the automotive domain, avionics, and control systems. Real-time systemsdiffer from traditional computer systems in their dependence on time asa correctness criteria, i.e., a late correct answer is useless for these systems.Embedded real-time systems today are more integrated, more parallel, andmore complex than ever before. In this thesis, we discuss limitations thataffect the applicability of real-time models, analysis methods, and schedulingapproaches to the realities of today's embedded systems and propose solutionsto address these challenges. We first look into the issue of shared resources and its effect on the mapping and scheduling of software tasks in a real-time system. Most task mappingapproaches proposed in the literature perform task mapping assuming independenttasks that do not share resources. Managing shared resources andtheir protection mechanisms is performed later. However, this approach mightrequire several rounds of iteration and can lead to inefficient results. In thisthesis, we explore the possibility of using different resource protection mechanismswithin a single system, and propose to tackle the design problem moreefficiently by jointly performing task allocation, scheduling, and resource protection mechanism selection. Two approaches are presented to solve this optimizationproblem: an optimal Mixed Integer Linear Programming (MILP)approach and an efficient heuristic. The proposed work is shown to significantly improve system schedulability. Experimental results indicate that theminimum utilization at which at least 95% of systems become scheulable canbe improved from 65%-70% for the best published task allocation algorithmsto 76%-85% using our heuristic with minimal memory cost. Even better resultscan be achieved using the MILP approach.Next, we look into the design of systems composed of components thathave different levels of criticality. Mixed-Criticality Systems (MCS) receivedmuch attention recently to due their industrial relevance. We focus on threechallenges in MCS design: task allocation, fault-tolerance, and model-baseddesign. For task allocation, we show that traditional task allocation algorithmscan be inefficient in a mixed-criticality context, and propose an alternative thatwe call dual-partitioned task allocation. Experiments show that for systemsthat have a utilization of 80% or higher, we can schedule 17% more systemson a given multicore platform using the dual-partitioned approach. Fault-tolerance is an important issue for MCS since these systems containa safety critical part. To design MCS that tolerate hardware transient faults,we propose a new mixed-criticality model that simultaneously addresses criticality,reliability, and Quality of Service (QoS). A schedulability test for thenew model is derived. Furthermore, to allow designers to incorporate the newmodel and analysis in their design process, we propose a design space explorationframework based on the new model that supports various fault-tolerancemechanisms. QoS improvements of up to 42.9% can be achieved using the newmodel compared to the traditional MCS model extended to support transientfaults.For model-based design, we propose algorithms to generate optimized semantic-preserving implementationsfor MCS specified using the SR model, with minimal functional delay addition.An optimal Branch-and-Bound based algorithm and an efficient heuristicare proposed for this purpose." --