In classification, and data mining in general, normally a sequential strategy is used, where successively a preparation, survey and modelling of data is executed. We propose a new paradigm, where no strict distinction is made between these phases, by which information can be gained during the modelling that can be used to improve the manipulation and survey process. For this a framework is presented, where classifiers based on distances van be combined in a hierarchical structure. This structure can be constructed interactively such that the data analyst can guide it, while the system makes suggestions, gives feedback and possibly takes decisions itself. To communicate the obtained information efficiently and to support the integrated process, some data visualisation techniques are developed specifically for this purpose.