In this work, we consider the resource allocation problem for task offloading from Internet of Things (IoT) devices, to a non-terrestrial network. The architecture considers clusters of IoT devices that offload their tasks to a dedicated unmanned aerial vehicle (UAV) serving as a multi-access edge computing (MEC) server, which can compute the task or further offload it to an available high-altitude platform station (HAPS) or to a lowearthorbit (LEO) satellite for remote computing. We formulate a problem that has as objective the minimization of the weighted sum delay of the tasks. Given the non-convex nature of the problem, and acknowledging that the complexity of the optimization algorithms impact their performance, we derive a lowcomplexity joint subchannel allocation and offloading decision algorithm with dynamic computing resource initialization, developed as a greedy heuristic based on convex optimization criteria. Simulations show the reduced delay obtained by including the different non-terrestrial nodes against architectures without them and state of the art benchmarks.