In general, even if the fibers of $\phi$ are connected and the map is proper, there need not be an almost complex structure on $B^{2k}$ such that the differential of $\phi$ is complex linear.  (I assume that you meant to assume that the kernel of the differential of $\phi$ is preserved by $J$, not that $J$ acts trivially on it, which doesn't make sense unless the kernel is $0$.)

Perhaps the most famous example of this phenomenon is the so-called 'twistor fibration':  $\pi:\mathbb{CP}^3\to \mathbb{HP}^1=S^4$.  The idea is that we regard $\mathbb{C}^4$ as $\mathbb{H}^2$ (where $\mathbb{H}$ is the ring of quaternions).  Then $\mathbb{CP}^3$ is the space of $1$-dimensional subspaces of $\mathbb{C}^4$ and the map $\pi$ is defined by letting $\pi(\mathbb{C}{\cdot}v) = \mathbb{H}{\cdot}v$ for every nonzero $v\in\mathbb{C}^4$.  The fibers of $\pi$ are complex lines in $\mathbb{CP}^3$, so they are holomorphic submanifolds of $\mathbb{CP}^3$ that are connected and compact, and $\pi$ is a submersion.  However, $S^4$ does not admit *any* almost-complex structure at all, let alone one for which the differential of $\pi$ is $\mathbb{C}$-linear.  In fact, it's not hard to see that this is a local failure, in the sense that, even after restricting $\pi$ to some (non-empty) open subset $U\subset\mathbb{CP}^3$, there still is no almost-complex structure on the image $\pi(U)$ for which the differential of $\pi$ is $\mathbb{C}$-linear.

There are many other examples, of course, even an example of a submersion $\phi:B^4\to B^2$, where $B^{2k}\subset\mathbb{C}^k$ is the usual open ball, such that the fibers of $\phi$ are connected complex curves in $B^4$ but there is no almost-complex structure on $B^2$ for which the differential of $\phi$ is $\mathbb{C}$-linear.  I can supply this example, too, if the OP is interested.

Meanwhile, if there is an almost complex structure on $B^{2k}$ for which the differential of $\phi$ is complex linear, then, yes, that almost complex structure has to be integrable.

The proof of this is straightforward:  Suppose that there is an almost-complex structure on $B^{2k}$ for which the differential of $\phi$ is $\mathbb{C}$-linear.  Let $\omega^1,\ldots,\omega^k$ be a $C^\infty(B^{2k})$ basis for the $(1,0)$-forms on $B^{2k}$ with respect to this almost-complex structure.  Then there will be functions $N^i_{\overline{p}\overline{q}}=-N^i_{\overline{p}\overline{q}}$ on $B^{2k}$ such that the equations
$$
\mathrm{d}\omega^i \equiv \tfrac12\,N^i_{\overline{p}\overline{q}}\,
\overline{\omega^p}\wedge\overline{\omega^q}
\quad\mathrm{mod}\quad \omega^1,\ldots,\omega^k
$$
for $1\le i\le k$ hold.  These functions vanish identically if and only if the almost-complex structure is integrable.  Now, let $^*\!\omega^i=\phi^*(\omega^i)$.  Because the differential of $\phi$ is $\mathbb{C}$-linear and $\phi$ is a submersion, these complex-valued $1$-forms are linearly independent and of type $(1,0)$ with respect to the $J$ on $B^{2n}$.  Thus, they can be completed to a $C^\infty(B^{2n})$-basis of $(1,0)$-forms for $J$ by choosing some $(1,0)$-forms $^*\!\omega^{k+1},\ldots,^*\!\omega^n$ on $B^{2n}$ such that  $^*\!\omega^1,\ldots,^*\!\omega^n$ are linearly independent.  However, because $J$ is integrable, we must have
$$
\mathrm{d}(^*\!\omega^j)\equiv 0\quad\mathrm{mod}\quad ^*\!\omega^1,\ldots,^*\!\omega^n
$$
for $1\le j\le n$.  This immediately implies that $\phi^*(N^i_{\overline{p}\overline{q}})=0$, and, since $\phi$ is a surjective submersion, this implies that $N^i_{\overline{p}\overline{q}} =0$, i.e., that the 'push-forward' almost-complex structure is integrable.