If $X$ and $Y$ are spaces then the Kunneth theorem implies that $$H_n(X \times Y;\mathbb{Z}) = \left(\bigoplus_{p+q = n}H_p(X;\mathbb{Z}) \otimes H_q(Y;\mathbb{Z})\right) \oplus \left(\bigoplus_{p+q=n-1} \operatorname{Tor}(H_p(X;\mathbb{Z}) ,H_q(Y;\mathbb{Z}))\right).$$
Now if $X$ and $Y$ are CW-complexes, then (correct me if I'm wrong) we know that $H_i(X;\mathbb{Z})$ and $H_i(Y;\mathbb{Z})$ are free abelian, which should imply that $\operatorname{Tor}(H_i(X;\mathbb{Z}) ,H_j(Y;\mathbb{Z})) = 0$ for any $i, j \geq 0$ and hence that $$H_n(X \times Y;\mathbb{Z}) = \bigoplus_{p+q = n}H_p(X;\mathbb{Z}) \otimes H_q(Y;\mathbb{Z}).$$
But now consider $K \times \mathbb{RP}^2$ (which is a CW-complex) where we let $K$ denote the Klein bottle, then in one calculation of $H_3(K \times \mathbb{RP}^2)$ I've seen, it was stated that $$H_3(K \times \mathbb{RP}^2) = \operatorname{Tor}(H_1(K;\mathbb{Z}), H_1(\mathbb{RP}^2;\mathbb{Z})) = \mathbb{Z}/2.$$
This seems to contradict what I've said above, is there some subtlety that I am missing?