In recent years, with the development of embedded software, the demand for context-aware software and the need for Context-Oriented Programming (COP) frameworks is increasing. In general, within the COP framework, each method is managed in units of layers, and time is required to activate the layers following context changes. Compared to the traditional single-thread method, the use of multi-threading to control the layer activation doubles the system efficiency. Moreover, embedded software usually includes mechanisms for exception handling, such as communication disconnection, mechanical or hardware failures, and insufficient energy. These mechanisms are expected to be types of high-priority operations, which also require multi-threading to achieve preemptive layer activation. In this study, we propose an architecture for preemptive layer activation known as PLAM by using multi-threading, which is based on the RTCOP framework. Using PLAM, the non-exception handling layers can be activated simultaneously, and when the context of the exception handling layer is perceived, the program can call a certain function to suspend or terminate the activation process of the other layers. Finally, we package a series of functions as APIs so that users can use these to customize the context listener. A test application confirms the feasibility of the proposed approach.