<p>Companies develop software in phases and launch them into multiple versions for early introduction of significant features while adding features and improving the reliability of the software with each release to beat the market competition. Software reliability is measured using Software Reliability Growth Models in terms of the expected number of failures in a given time interval. Depending upon the severity of the faults, software faults can have different impacts on the systems. Risk-associated with faults in the software need to be addressed on a priority basis to maintain safety, customer trust and cost-effectiveness. Hence, a multi-version software reliability model is proposed, considering impact-based fault severity classification and imperfect debugging. Also, as the software testing progresses for each version, the fault detection rate changes due to factors like changing testing environment, testing strategy, skill, testing personnel etc. Therefore, the failure dataset of each software version is uniquely studied to understand the failure pattern. The failure patterns are captured through best-suited distribution functions to predict the expected number of failures more precisely. The proposed model is compared with the existing models and found to be providing more accurate results on two datasets, one of which is taken from literature while another is a masked dataset prepared from testing data of a product.</p>