Deep Neural Networks (DNNs) and Generative AI (GenAI) models are widely used across various domains, including healthcare, where their training and inference require substantial computational resources. Cloud infrastructures facilitate multi-tenant execution of machine learning workloads, particularly on hardware accelerators like GPUs, to maximize resource utilization, reduce costs, and improve accessibility for users with varying computational demands. While existing scheduling strategies optimize job completion time, resource allocation, and latency reduction, they often overlook privacy concerns, which are critical for sensitive healthcare applications. In this chapter, we introduce a privacy-preserving scheduling framework that ensures secure execution of ML workloads while maintaining efficiency in multi-tenant cloud environments. Additionally, inference in GenAI models, such as Stable Diffusion, differs from traditional DNNs, necessitating iterative refinement rather than a single forward pass. This chapter presents a novel, privacy-preserving, low-latency inference framework tailored for Stable Diffusion as a Service, addressing the unique challenges of deploying GenAI models in cloud environments.