A cross-layer architecture that achieves significant power savings, while enhancing the end-to-end QoS provisioning and granularity in wireless ad hoc networks is proposed in this thesis. Recently, a new concept called service vector has been introduced, which enables an end host to choose different service classes along its data path. This scheme enhances the user benefits from the network services and network resource utilization, while maintaining the simplicity and scalability of the current Differentiated Services (DiffServ) network architecture. This thesis explores the application of this concept on wireless ad hoc networks and provides a cross-layer architecture based on the combination of delay-bounded wireless link level scheduling and the network layer service vector concept, which enables a wireless ad hoc network to achieve significant power savings and finer end-to-end QoS granularity. The impact of various traffic arrival distributions and flows with different QoS requirements on the performance of this cross-layer architecture is also investigated and evaluated.