<p>With the growing demand for scalable and efficient distributed systems, microservices and serverless architectures have become key enablers for modern applications. This study focuses on designing a microservices-based image classification workflow that integrates an ensemble learning algorithm to achieve robust and accurate predictions. The workflow is modeled using Extended Queueing Networks, enabling detailed representation of function stations and their associated queues. To assess system performance and ensure Quality of Service, key metrics are defined, and the model is analyzed using Java Modeling Tools. Performance analysis at both the station and system levels highlight critical bottlenecks and inform a capacity planning strategy to ensure stability under varying workloads. The results show that the proposed capacity planning approach can avoid overutilization by dynamically increasing the number of function instances, thus maintaining the utilization of all resources within a predefined range. This research provides valuable insights into designing reliable, scalable, and efficient workflows on Function-as-a-Service platforms, addressing the challenges of queue management and system overutilization in serverless environments. </p>