Electromechanical compact axes, which integrate electric motors with screw drives, are becoming increasingly popular in decentralized drive systems due to their high power density and precise control. These systems are emerging as effective alternatives to hydraulic systems, driven by the growing demand for more compact and efficient drive solutions. However, the compact architecture of electro-cylinders poses critical thermal management challenges. High power density and the proximity of heat-generating components within the design often result in heat accumulation, leading to various performance degradations. Limited pathways for heat dissipation create the risk of thermal buildup, which can degrade motion accuracy through thermo-elastic effects and compromise long-term reliability as heat-related stress affects components such as bearings and nuts. Predicting these effects with current simulation techniques is possible but comes at a high computational cost or accuracy loss, making design optimizations challenging and highlighting the essential role of Model order reduction (MOR) in improving simulation efficiency and enabling faster, more cost-effective design analysis without sacrificing accuracy. While MOR can reduce computation time, its integration into thermal Finite Element Method (FEM) models is not yet standardized. This requires addressing non-linearities arising from heat generation patterns, specifically mechanical friction, electrical resistance, and heat radiation, since conductivity and specific heat capacity remain constant across the temperature range considered. The challenge intensifies in mobile systems with continuously changing convective and conductive heat transfer conditions. To address these aspects, this paper presents a framework that integrates MOR into thermal FEM models, accounting for non-linear thermal behaviors and enabling efficient, accurate simulations across diverse operational scenarios. This framework combines MOR in Matlab with Ansys-based subsystem discretization, focusing on active interfaces and matrix reduction of conductivity and capacity. Mobile components are discretized at critical zones to accurately model dynamic boundary conditions. Two thermal models are developed: a non-reduced model for high fidelity and a reduced model for faster simulations that maintain accuracy at key interfaces. This approach is particularly suited for mobile systems with dynamic thermal conditions, offering scalable, real-time analysis for electromechanical systems. Experimental validation was performed on a grease lubricated compact axle, at varying travel speeds to evaluate the performance in different operating regimes. The strong correlation between model predictions and experimental results underscores the importance of fast and high-accuracy thermal simulations enabled by MOR techniques. This approach enhances the reliability and operational efficiency of electro-mechanical axles, supporting their potential for broader industrial adoption.