As embedded systems become increasingly complex, not only are dependability and timeliness critical quality indicators, but also their ability to dynamically adapt to changes in their runtime environment. Service-oriented Architecture (SOA) offers a potential solution to this challenge by allowing services to be dynamically bound at runtime. However, resource contention between services and resource saturation can result in significant Quality of Service (QoS) degradation. To address these issues, this paper proposes a runtime, resource-aware architecture that combines resource monitoring with dynamic workflow orchestration to mediate resource contention within the orchestration environment. The architecture was evaluated on a medium-size service-oriented case study, deployed on a resource-constrained device.