This paper proposes the design of an evolutionary algorithm for building classifiers specifically aimed towards performing classification and sentiment analysis over texts. Moreover, it has properties taken from Artificial Immune Systems, as it tries to resemble biological systems since they are able to discriminate harmful from innocuous bodies (in this case, the analogy could be established with negative and positive texts respectively). A framework, namely OpinAIS, is developed around the evolutionary algorithm, which makes it possible to distribute it as an open-source tool, which enables the scientific community both to extend it and improve it. The framework is evaluated with two different public datasets, the first involving voting records for the US Congress and the second consisting in a Twitter corpus with tweets about different technology brands, which can be polarized either towards positive or negative feelings; comparing the results with alternative machine learning techniques and concluding with encouraging results. Additionally, as the framework is publicly available for download, researchers can replicate the experiments from this paper or propose new ones.