Over the last decade, immersive technologies have grown significantly, offering users highly interactive and increasingly dynamic environments. Head-mounted displays are currently among the leading devices that enable access to high-quality, realistic immersive experiences. However, these technologies, due to their nature, often present several limitations in terms of storage capacity and response time. Due to these limitations, increasingly sophisticated software architectures are required to ensure more fluid and efficient user experiences and provide real-time responses to all demands. For this reason, the ability to support parallel processing for complex tasks such as data inference or the incorporation of machine learning (ML) models has become a critical challenge. Recently, our team published a general architecture, composed of six interrelated components, that allows the grouping and organization of a diverse set of ML models or data inference methods adaptable for immersive environments. In this paper, we address initial ideas for an efficient implementation of the fourth component, “Task Orchestration and Scheduling,” of this architecture. This component manages the concurrent execution of multiple models based on extracted features and user interactions. The objective of this work is to explore the role of software architectures in improving the multitasking experience of users interacting with immersive applications. We focus specifically on task scheduling, parallelism, and scalability. Furthermore, we provide interesting elements for efficient code development for parallel programming and show promising results comparing sequential and parallel programming approaches.