When designing reliability-aware digital circuits, either hardware or software techniques may be adopted to provide a certain degree of failure detection/tolerance, caused by either hardware faults or soft-errors. These techniques are quite well established when working at a low abstraction level, whereas are currently under investigation when moving to higher abstraction levels, in order to cope with the increasing complexity of the systems being designed. This paper presents a model of soft error effects to be adopted when defining software-only techniques to achieve fault detection capabilities. The work identifies on a generic IP processor the misbehaviors caused by soft errors, classifies and analyzes them with respect to the possibility of detecting them by means of previously published approaches. An experimental validation of the proposed model is carried out on the Leon2 processor.