Quality of service (QoS) becomes an important issue as systems become more open and therefore less predictable. In such contexts, it is difficult, using static approaches, to ensure that a system will provide the expected QoS. In dynamic approaches, services are adapted to provide the best QoS according to the execution context. This article describes an object-based framework to deal with QoS in domains in which temporal faults are a major QoS criterion. The QoS optimization policy influences most of the architectural design decisions. The implementation is based on meta-objects that collectively manage the QoS and form a two-level decision framework: at the global level, objects share the same decision policy, aiming to reduce timing faults, while at the object level, any specific QoS criteria may be applied. The QoS description is made using UML extension mechanisms. Finally, the main phases of QoS management are detailed: prediction, establishment and operation, including observation and negotiation.