Virtual reality applications offer the promise to immerse end users in a synthetic environment where several actions could be observed, simulated, and reproduced, before transferring them to reality, which makes them particularly appropriate for training. Yet, when the training requires complex handling of information, the tasks become cognitively intensive, and developing adequate applications becomes challenging. To address this challenge, we define a method for developing head-mounted-display-based virtual reality applications for modular training tasks, composed of a training model with parameters, a step-wise approach for supporting this development, and a software framework enacting the application of this approach. Authoring such applications is expected to become more flexible and provide personalization facilities. To evaluate the impact of this method, we define a case study concerning an application for training school teachers who deal with a variety of situations in a classroom for an experiment involving N= 7 participants for a set of tasks. Pre-study and post-study acceptances reveal the impact of the software framework and a workload evaluation is conducted using the NASA TLX questionnaire.