Sensor data evaluation and environment modeling represent a central part in the design of robotic systems. The environment modeling process usually is done by several data processing modules by means of sensor data fusion, combination of a priori knowledge with currently sensed data, etc. Depending on the application scenario there is usually a trade-off between accuracy and time consumption of the environment modeling process. Some domains require a quite accurate environment model and processing time plays a secondary role. More reactive domains, however, demand for rather short cycle times, even if this results in a less accurate model. For some applications the required degree of accuracy may vary depending on the current situation or task to perform and some parts of the sensed information may be more important than others. Several architectures were proposed to address this problem by special mechanisms which schedule the available resources according to the needs of the current tasks which is usually denominated by the term ``selective'' or ``focused'' attention. In our work we will show how these concepts can be put into practice by a flexible and open software design which is applied for our CoPs RoboCup midsize team.