Ambient assisted living technologies (AALT) are considered to offer means for reducing health care burden to societies by prolonging capability for independent living and by improving the quality of life for elderly persons suffering from dementia and their caregivers. This research presents a novel activity support framework for transforming cooking activities in to workflow process models optimized for people suffering from dementia. This is achieved by first splitting a cooking activity in to tasks composed of set of basic actions. Due to nature of dementia, actions taken by a care recipient can be considered to be stochastic. This research presents a model of a care recipient suffering from dementia based on data from literature. The model contains probability distribution of possible outcomes mapped to available basic actions in a cooking activity. This allows the use of Markov decision process (MDP) to determine optimal sequence of actions to perform a cooking activity. The resulting process models can be used as set of instructions by a prompting system or as a process model for a belief monitor type of AALTs. The designed framework was tested by comparing manually constructed and verified process model to one produced by a test computer program. Further research is needed to refine the care recipient model, study transforming cooking recipes into cooking activity models and to validate the framework in real-world settings.