The usual proof of Kodaira's lemma should work:
Let $D$ be $f$-big and let $A$ and arbitrary relatively ample and effective divisor. Then consider the short exact sequence:
$$
0\to \mathscr O_X(mD-A) \to \mathscr O_X(mD) \to \mathscr O_X(mD)\left|_A\right. \to 0
$$
In the absolute case, one observes that as $m\to\infty$, the dimension of $H^0(X, \mathscr O_X(mD))$ grows as $m^n$ where $n=\dim X$ while the dimension of $H^0(A, \mathscr O_X(mD)\left|_A\right.)$ can't grow faster than $m^{n-1}$, so for $m\gg 0$ the induced map cannot be injective and hence $H^0(X,\mathscr O_X(mD-A))\neq 0$ and we're done.
For the relative case one can do the same thing, just take $f_*$ instead of $H^0$. We have the exact
$$
0\to f_*\mathscr O_X(mD-A) \to f_*\mathscr O_X(mD) \to f_*\mathscr O_X(mD)\left|_A\right.
$$
and we may observe that the rank of $f_*\mathscr O_X(mD)$ and that of $f_*\mathscr O_X(mD)\left|_A\right.$ can be computed as $H^0$ on the general fiber. We get the same conclusion, that is, that the map $f_*\mathscr O_X(mD) \to f_*\mathscr O_X(mD)\left|_A\right.$ can't be injective and hence $f_*\mathscr O_X(mD-A)\neq 0$.
Remark: I suppose one reason why one might want to take $Y$ to be affine is that in that case $f_*$ is given by an $H^0$ and hence $f$-effective actually implies effective.
Addition, to answer the extra question in the comments.
In the situation as described in the question, assume that $Y$ is quasi-projective and let $H$ be a very ample effective divisor on $Y$. Further assume that $H$ is chosen so that $A=D+f^*H$ is big on $X$ (i.e., not just $f$-big). In particular, we may assume that $A$ is effective. Letting $E=f^*H$ shows that the required decomposition is possible. If $H$ is chosen generally in its linear system, then one may even assume that $E$ does not have any exceptional components.