Embedded systems executing soft real-time workloads are increasingly expected to maintain predictable timing behaviour under varying computational load. When a task repeatedly exceeds its deadline on a resource constrained local platform, system quality degrades and recovery is not guaranteed. Edge computing, which places additional processing resources close to the device, offers a potential remedy by allowing tasks to be offloaded to a faster remote server. However, the communication cost of offloading means that migration is only beneficial under certain conditions and an uninformed decision can worsen timing rather than improve it.This thesis investigates whether a lightweight runtime mechanism can reliably detect deadline violations and make correct, network-aware offloading decisions for soft real-time tasks. Two components are designed and implemented, Task Runtime Execution Monitor (TREM), monitoring agent that detects per-job deadline violations at runtime and Network-aware Offloading Decision Engine (NODE), decision engine that estimates whether executing a violating task on the edge server would allow it to meet its deadline given the current network conditions.The mechanism is evaluated through four experiments that vary the presence of the decision engine, the edge speed factor and the network quality. The results show that TREM correctly identifies persistent violation states, that NODE approves offloading only when it is genuinely beneficial and that the mechanism eliminates deadline violations once triggered, a task that misses every one of 37 consecutive deadlines without the mechanism misses only 6 before NODE offloads it and none of the following 35 job instances violate their deadline. The work demonstrates that dynamic, network-aware offloading is a viable strategy for recovering soft real-time tasks from persistent deadline violation states on resource-constrained platforms.