The thermal management of electric motors is critical to maintain high performance, efficiency, and durability. This paper presents a novel multiphysics reduced order modeling (ROM) approach that integrates seamlessly into the design and development cycle of electric motor, covering component predesign, integrated system analysis, and verification-and-validation stages. During the early design phase, 2D finite element (FE) simulations using GT-SUITE are employed to evaluate losses across various operating points. Subsequently, calibrated thermal-node network models are derived to enable the analysis of different cooling concepts. This pre-design analysis provides valuable insights into thermal behavior, allowing for informed design decisions. For later development stages, a comprehensive thermal mass network based on actual geometry is developed to perform system-level thermal analyses. To facilitate faster design optimization, control calibration and hardware testing, a neural-network based metamodeling approach is also explored in this paper by generating metamodels from high-fidelity multiphysics models to integrate seamlessly into X-in-the-Loop (XiL) simulations. In conclusion, the presented methodologies and toolset not only accelerates the development process but also enhances the accuracy of thermal predictions throughout the design and validation phases by bridging the gap between early design simulations and real-world system integration.