Along with the rapid growth of IoT technologies and devices, their solutions are currently being applied on various domains such as health-care, transportation and agriculture, but mainly on crowd monitoring and emergency handling. The latter is a safety critical IoT system based on collecting and analyzing the real-time data to perform proper actuation. In order to engineer such a high quality IoT application, a proper software architecture should be designed. In order for the software architecture to be able to optimize critical requirements such as fault-tolerance, performance and energy consumption, it ought to: 0 adapt itself to real-time environment transformation, ii) be designed in a proper level of elements distribution. In this paper, we critically analyze a set of IoT distribution and self-adaptation patterns to identify their suitable architectural combinations. Further, we use our IoT modeling framework (CAPS) to model an emergency handling system. Based on these, we design two quality driven architectures to be used for a forest monitoring and evacuation example and qualitatively evaluate and compare them.