I'm not sure you understand how bra-ket notation is used. You should not have x as a parameter in your bras and kets unless you mean a wave-packet rather than a particular eigenstate. I believe you are asking about.
$$\int dx e^{-ikx}u^*_{nk}(x) \frac{d}{dx} e^{ik'x}u_{n'k'}(x) $$ $$= \int dx e^{-i(k-k'
)x}( -iku^*_{nk}(x)u_{n'k'}(x) + u^*_{nk}(x)\frac{d}{dx} u_{n'k'}(x) ) $$
I think we get something like ##-ik\delta(k-k')\delta_{nn'}## for the first term, but I can't think of any reason why we would get anything in particular for the second term - it depends entirely on the unspecified nature of the ##u_{nk}(x)## functions.