In mobile cloud computing, traditional frameworks often treat mobile devices as passive data collectors, relying heavily on centralized cloud centers for model processing. This centralized approach limits data-processing accuracy and efficiency due to bandwidth constraints, communication latency, and limited computational utilization of mobile devices. Furthermore, as mobile applications increasingly require real-time, personalized services, these limitations restrict model adaptability and responsiveness, emphasizing the need for a distributed approach that can dynamically adjust to varying service demands and network conditions. This paper proposes FLOM (Framework-Level Self-evolution Optimization Model), a novel framework designed to support autonomous, adaptive model optimization across multi-terminal systems. Unlike traditional methods, FLOM leverages the computational capacity of mobile devices to enable on-site model updates and intelligent data allocation through a collaborative network of cloud and mobile nodes. FLOM incorporates deep reinforcement learning for dynamic model evolution, allowing real-time adaptation across terminals. By utilizing model information from each device, FLOM achieves balanced self-evolution, optimizing processing accuracy and system resilience. The framework introduces three major innovations: (1) Self-adaptive Model Evolution by utilizing mobile terminal resources to locally update models, FLOM reduces dependence on centralized data centers, ensuring faster, context-aware adjustments that enhance real-time service delivery; (2) Intelligent Resource Allocation supporting real-time, task-targeted distribution to enhance fault tolerance and system efficiency, FLOM effectively allocates resources to handle diverse service needs across terminals; and (3) Collaborative Optimization by integrating model characteristics from multiple terminals, FLOM enables unified, system-wide optimization that improves model robustness and adaptability. Experimental results show that FLOM significantly improves data accuracy, reduces processing time, and enhances fault tolerance compared to traditional frameworks, making it a robust and versatile solution for real-time, adaptive mobile cloud applications.