The edge architecture introduces challenges due to the presence of heterogeneous constrained devices and the dynamic nature of the network. Dynamic service orchestration is essential for efficient utilization of network resources in this context. Service orchestration plays a critical role in 5G networks in automating end-to-end service deployment and operations, and 5G network slicing. However, existing approaches on service orchestration primarily focus on resource availability, lacking a comprehensive understanding of network conditions-cross network layer orchestration. As a result, providing efficient service orchestration for delay-sensitive services remains a significant research area. To address this gap, this paper proposes a cognitive service orchestration framework that leverages not only application-level resource demands, but also the real-time status of the network infrastructure. This cognitive framework incorporates Reinforcement Learning, enabling it to dynamically interact with the network environment and continuously update policies for intelligent decision-making in complex 5G networks. The efficacy of the proposed framework is evaluated using an object detection application in smart cities. The evaluation results show that the proposed framework achieves a significant reduction in latency as compared to OpenELB, with a remarkable 58% decrease, which highlights its efficiency and effectiveness in meeting the requirements of delay-sensitive services.