Digital Twins (DTs) are complex systems that increasingly support critical applications across various domains, including agriculture, transportation, home automation, automotive, aerospace, and healthcare. Essentially, a DT is a virtual representation of a physical or digital object, process, or system, serving as a bridge between the physical and digital worlds. They function by enabling real-time monitoring, simulation, and optimisation, offering significant advantages: 1) Simulation and Testing: DTs facilitate real-time monitoring and simulation of their physical or digital counterparts, allowing engineers and developers to test scenarios, predict outcomes, and optimise performance without risking the actual system. 2) Data Integration: By incorporating data from sensors and IoT devices, DTs provide a complete view of system performance in real-time, enabling early issue detection and informed decision-making. 3) Predictive Maintenance: In industries such as manufacturing, aerospace, and healthcare, DTs anticipate potential failures before they occur, enabling proactive maintenance, reducing downtime, and minimising costs. 4) Enhanced Decision-Making: DTs offer stakeholders a clear, data-driven understanding of operations, leading to improved planning, design, and management across multiple sectors. Embedded systems, particularly Cyber-Physical Systems (CPS), are characterised by the tight integration of computational (cyber) and physical components. In this thesis, these CPS are referred to as the actual systems. Digital twin revolutionise our interactions and management of the actual system, fostering innovation and efficiency. The design and testing of DT systems need simulation models, which provide engineers with the means to analyse system behavior and ensure compliance with performance requirements. However, the complexity of the actual system and DTs translates into computationally expensive models, posing significant challenges when real-time constraints necessitate in time decision-making with strict requirements for timeliness, and accuracy. To address these challenges, this thesis introduces the SACube framework, a novel approach for dynamically switching between high-validity model-the model with broader validity frame- and less-detailed models–models with narrower validity frame to achieve computational efficiency withoutcompromising accuracy. This framework tackles critical questions of how, when, and where these switching happens. Moreover, we emphasise the importance of the model validity region, which is as essential as the model itself. Sometimes the model validity is available upfront and already given, but sometimes not. To address this, we introduce the Decision-Centric Technique to determine a model’s validity within the context of a decision-maker and store this information in our framework. We further demonstrate the application of this framework in a DT system with real-time requirements, providing detailed workflows. The proposed methods in this thesis are assessed using two case studies: an automated highway lane change maneuver system and a DT of a traffic system. These use cases illustrate the actual application of adaptive approaches to fulfil real-time requirements while preserving the accuracy and usefulness of predictive models. This thesis presents a structured approach for optimising multi-modal system simulations, particularly in computationally intensive domains like urban traffic networks. By defining a validity frame for surrogate models, it ensures seamless mode switching while maintaining accuracy and minimising redundancy. The proposed ensemble of adaptive surrogate models enhances performance modeling, enabling efficient analysis of dynamic, multi-modal systems. This thesis lays the foundation for scalable and robust modelling of complex systems through the integration of adaptive abstraction and approximation, machine learning, and surrogate modelling techniques. The suggested methodologies are substantiated by thorough experimentation, demonstrating their applicability across various domains and enhancing the cutting-edge in modelling and simulation for CPS and DTs.