Real-time embedded systems are found in various application domains. They have to offer an increasing number of functionalities and to provide the highest Quality of Service despite possible failures due to faults or processing overloads. In such systems, programs are characterized by upper bounds on finishing times and the QoS is assessed by the ratio of successful deadlines. In this thesis, we deal with this issue. We focus on a uniprocessor architecture in the framework of a firm real-time application that accepts deadline missing under some specified limits. Tasks are assumed to be periodic. Our first contribution lies in the proposition of a novel model for tasks which is called BGW model. It is drawn from two approaches respectively known as the skip-over model and the Deadline Mechanism. The first one provides timing fault-tolerance through passive dynamic software redundancy with two versions. The second one copes with transient processing overloads by discarding instances of the periodic tasks in a controlled and pre-specified way. We give a feasibility test for this model. In a second part, we describe the behavior of dynamic priority schedulers based on EDF (Earliest Deadline First) for BGW task sets. A performance analysis is reported which is mainly related to QoS evaluation and measurement of overheads (complexity of the scheduler). The following contribution concerns more sophisticated schedulers that permit to enhance the QoS as to improve service balancing.