A class of nonparametric estimates for errors-in-variables models over the sets of sufficiently smooth functions are proposed. Total least squares penalization is used.