Performance bugs may greatly affect the quality of the system being developed and even cause irreversible damage in some critical sectors. Hence profiling - one of the currently most widespread technique of performance analysis - is usually applied to find the bugs. However, most of the current solutions commonly lack comprehensible graphical outputs and detailed analysis of algorithms in regard to their complexity. This thesis introduces a novel profiling tool which focuses on automatic estimation of complexity of dynamic data structures. The proposed approach collects statistical data out of program runs and uses regression analysis to find the most accurate model serving as an estimate of algorithmic complexity. The resulting prototype was subjected to a series of experiments that evaluate the accuracy of the results, demonstrate practical uses and illustrate the graphical output of the tool.